HP's Chandrakant Patel Gives Advice for Engineers: Stick to Your Fundamentals

Not only is it an essential part of the process today, in the cyber age or the data-centric age, it is a central part of the process because we need to design energy-efficient data centers. That's where fundamentals come in. But not only is it true today, but indeed going into the future, fundamentals will be central.

The 20th century was about the Machine Age, about fundamentals. The latter part of the 20th century was the cyber age, the Information Age, built on data centers. The data centers were built on fundamentals. Once they were built, we somehow got into this thing of, 'All the data is available, I don't need to know the first thing. I don't need to know the second law of thermodynamics. It's all there.' That is unfortunate because if you look at the 21st century, it's the integration of cyber and physical. It's the Machine Age together with the Digital Age. The cyber-physical age necessitates depth in fundamentals.

Unfortunately, the black box we have is not that simple. You must have domain knowledge. In fact, I wrote an article recently on LinkedIn, a blog which I called 'Machine Learning Requires Domain Knowledge.' That shows that if you just took data and tried to figure out what is happening, you could see some phenomena occurring and you could try to correlate it. If you don't have domain knowledge, you might say, 'This happened because of the following thing.' You have a hypothesis, you prove or disprove it. You can't get to causation. In my article, I wrote about failures of disk drives in a data center. Data was there. Then I looked at the rotation of the blades: four blades, 15,000 RPM, 250 Hertz, times four is 1,000 Hertz of frequencies. So from dynamics of structures I deduced that the fan rotational speed is causing the arm to vibrate, which doesn't cause a failure but has throughput problems. People take drives out, send it back to the supplier. 'No error found.' It comes back. Warranty cost, energy cost, all because we did not look at what was happening from a fundamentals point of view. That is a classic example where, like I share here, machine learning comes together with domain knowledge. That's a simple example, a simple resonance issue, just like a box girder bridge. The Tacoma Narrows Bridge is an example.

The first lady of songs, yeah, song was played back; the glass would shatter. Yes, it was a prime-time commercial. People knew what was resonance. That was an age of fundamentals. Today we are in the same thing. We are building large systemic systems like the data center. When these phenomena are occurring, if we are just going to guess at the phenomena, we will be inefficient. So even in the simple case, I need to get qualified domain knowledge. So after I wrote that article, one of the professors from Virginia Tech, a channel grant, said, 'You're absolutely right. In AI, people used to call it codifying domain theories.' In other words, what you do is the people who know the fundamentals, they codify it. Maybe it goes on engineering comm where you codify a whole section so that people who don't... heat transfer phenomena, this is very complex. There you need people who have, say, a master's in mechanical engineering and a PhD in computer science. So you need depth in almost two fields to work. So there are various levels where we will have a combination of fundamentals and data science come together. So it is definitely the age of fundamentals. That's how I trained all three of my kids. They were also engineers. They have to start from a fundamentals perspective.